Unverified50% confidenceFactExact time
The guide notes that each task should be evaluated more than once, because some regression problems appear in only one of two runs, and a single-run eval will miss them.
1
Sources
50%
Confidence
Long-term
Relevance
10/24/2026
First Seen
Sources
Related Entities
Related Claims
Unverified异步HITL模式面临幂等性保证挑战，系统需通过唯一任务ID和状态机校验防止重复执行71% similarUnverifiedA single execution-type agent exhibits three typical problems during long-running tasks: goal drift, shortcut-taking, and mid-task cessation70% similarUnverifiedIn long-running tasks spanning several days, a single API call failure can break the entire task chain, making error recovery and retry mechanisms critically important.68% similarUnverifiedIdempotency keys only solve the duplicate-request problem, not the billing problem after a multi-step task fails, because they are bound to a single request rather than the whole task.68% similarUnverifiedWhen running Agent tasks on shared hardware, common failure causes include the workspace becoming read-only, clock drift causing tokens to expire, external hosts being blocked, and two tasks writing to the same directory at the same time; what these problems have in common is that the test is whether the default environment state remains unchanged, not machine capability.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1092546API
curl https://kongchang.com/api/v1/knowledge/claims/1092546MCP
get_claim(id=1092546)